import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
from tensorflow.keras import regularizers
from tensorflow.keras.layers import *


def AUTOMAP_Branch(config):

    fc_1 = keras.Input(shape=(config.im_h,config.im_w,3), name='input')
    fc_1b=layers.Reshape([config.im_h*config.im_w*3])(fc_1)
    #with tf.device('/gpu:0'):
    fc_2 = layers.Dense(config.fc_hidden_dim, activation='tanh')(fc_1b)
   # with tf.device('/gpu:1'):
    fc_3 = layers.Dense(config.fc_output_dim, activation='tanh')(fc_2)

    fc_3 = layers.Reshape((config.im_h,config.im_w,2))(fc_3)
    
    #fc_3 = layers.ZeroPadding2D(4)(fc_3)
    
    #Jx branch
    
    c_1a = layers.Conv2D(64,5,strides=1,padding='same',activation='relu')(fc_3)
    c_2a = layers.Conv2D(64,5,strides=1,padding='same',activation='relu')(c_1a)
    
    c_3a = layers.Conv2DTranspose(1,7,strides=1,padding='same')(c_2a)
    
    
    
    #Jy branch 
    c_1b = layers.Conv2D(64,5,strides=1,padding='same',activation='relu')(fc_3)
    c_2b = layers.Conv2D(64,5,strides=1,padding='same',activation='relu')(c_1b)
    
    c_3b = layers.Conv2DTranspose(1,7,strides=1,padding='same')(c_2b)
    
    #o2 = layers.Reshape(config.im_h*config.im_w)(c_3b)
    output=[c_2a,c_2b,c_3a,c_3b]
                        
    model = keras.Model(inputs = fc_1,outputs = output, name='output')
    model.summary()
    return model

def UNET_Model_5(config):
    inputs = keras.Input(shape=(config.im_h,config.im_w,3), name='input')

    #with tf.device('/GPU:0'):
#     with tf.device(devices[0].name):
    
    conv1 = Conv2D(128, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(inputs)
    conv1 = Conv2D(128, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(conv1)
    pool1 = AveragePooling2D(pool_size=(2, 2))(conv1)
    
    conv2 = Conv2D(256, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(pool1)
    conv2 = Conv2D(256, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(conv2)
    pool2 = AveragePooling2D(pool_size=(2, 2))(conv2)
    
    conv3 = Conv2D(512, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(pool2)
    conv3 = Conv2D(512, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(conv3)
    pool3 = AveragePooling2D(pool_size=(2, 2))(conv3)
    
    conv4 = Conv2D(1024, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(pool3)
    conv4 = Conv2D(1024, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(conv4)
    drop4 = Dropout(0.5)(conv4)
    pool4 = AveragePooling2D(pool_size=(2, 2))(drop4)
    
    conv5 = Conv2D(2048, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(pool4)
    conv5 = Conv2D(2048, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(conv5)
    drop5 = Dropout(0.5)(conv5)

#     with tf.device('/gpu:1'):

    up6 = Conv2D(1024, 2, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(UpSampling2D(size = (2,2))(drop5))
    merge6 = concatenate([drop4,up6], axis = 3)
    conv6 = Conv2D(1024, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(merge6)
    conv6 = Conv2D(1024, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(conv6)
    
    up7 = Conv2D(512, 2, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(UpSampling2D(size = (2,2))(conv6))
    merge7 = concatenate([conv3,up7], axis = 3)
    conv7 = Conv2D(512, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(merge7)
    conv7 = Conv2D(512, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(conv7)
    
    up8 = Conv2D(256, 2, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(UpSampling2D(size = (2,2))(conv7))
    merge8 = concatenate([conv2,up8], axis = 3)
    conv8 = Conv2D(256, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(merge8)
    conv8 = Conv2D(256, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(conv8)
    
    up9 = Conv2D(128, 2, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(UpSampling2D(size = (2,2))(conv8))
    merge9 = concatenate([conv1,up9], axis = 3)
    conv9 = Conv2D(128, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(merge9)
    conv9 = Conv2D(64, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(conv9)
    conv9 = Conv2D(2, 3, activation = 'tanh', padding = 'same', kernel_initializer = 'he_normal')(conv9)
#         output = Conv2D(2, 1, activation = 'linear')(conv9)
    output = Conv2D(2, 1, activation = 'tanh')(conv9)

    model = keras.Model(inputs = inputs, outputs = output, name='output')
    model.summary()
    return model